CDs are recurring units in polypeptide chains (sequence and structure motifs), the extents of which can be determined by comparative analysis. Molecular evolution uses such domains as building blocks and these may be recombined in different arrangements to make different proteins with different functions. The CDD contains sequence alignments that define the features that are conserved within each domain family. Therefore, the CDD serves as a classification resource that groups proteins based on the presence of these predefined domains. CDD entries often name the domain family and describe the role of conserved residues in binding or catalysis. Conserved domains are displayed in MMDB Structure summaries and link to a sequence alignment showing other proteins in which the domain is conserved, which may provide clues on protein function.

The collections of domain alignments in the CDD are imported either from two databases outside of the NCBI, named Pfam (8) and SMART (9); from the NCBI COB database; from another NCBI collection named LOAD; and from a database curated by the CDD staff. The first task is to identify the underlying sequences in each collection and then link these sequences to the corresponding ones in Entrez. If the CDD staff cannot find the Accession numbers for the sequences in the records from the source databases, they locate appropriate sequences using BLAST. Particular attention is paid to any resulting match that is linked to a structure record in MMDB, and the staff substitute alignment rows with such sequences whenever possible. After the staff imports a collection, they then choose a sequence that best represents the family. Whenever possible, the staff chooses a representative that has a structure record in MMDB.

Once imported and constructed, each domain alignment in CDD is used to calculate a model sequence, called a consensus sequence, for each CD. The consensus sequence lists the most frequently found residue in each position in the alignment; however, for a sequence position to be included in the consensus sequence, it must be present in at least 50% of the aligned sequences. Aligned columns covered by the consensus sequence are then used to calculate a PSSM, which memorizes the degree to which particular residues are conserved at each position in the sequence. Once calculated, the PSSM is stored with the alignment and becomes part of the CDD. The RPS-BLAST tool locates CDs within a query sequence by searching against this database of PSSMs.

RPS-BLAST (Chapter 16) is a variant of the popular Position-specific Iterated BLAST (PSI-BLAST) program. PSI-BLAST finds sequences similar to the query and uses the resulting alignments to build a PSSM for the query. With this PSSM the database is scanned again to draw in more hits and further refine the scoring model. RPS-BLAST uses a query sequence to search a database of precalculated PSSMs and report significant hits in a single pass. The role of the PSSM has changed from "query" to "subject"; hence, the term "reverse" in RPS-BLAST. RPS-BLAST is the search tool used in the CD-Search service.

Analogous to the Structure Summary page, the CD Summary page displays the available information about a given CD and offers various links for either viewing the CD alignment or initiating further searches (Figure 4). The CD Summary page can be retrieved by selecting the CD name on any page.

In 2002, NCBI released the first group of curated CD records, a new and expanding set of annotated protein multiple sequence alignments and corresponding structure alignments. These new records have Accession numbers beginning with "cd" and have been added to the default CD-Search database. Most curated CD records are based on existing family descriptions from SMART and Pfam, but the alignments may have been revised extensively by quantitatively using three-dimensional structures and by re-examining the domain extent. In addition, CDD curators annotate conserved functional residues, ligands, and co-factors contained within the structures. They also record evidence for these sites as pointers to relevant literature or to three-dimensional structures exemplifying their properties. These annotations may be viewed using Cn3D and thus provide a direct way of visualizing functional properties of a protein domain in the context of its three-dimensional structure. (See Box 3 and Figure 7.)

The term "domain" refers in general to a distinct functional and/or structural unit of a protein. Each polypeptide chain in MMDB is analyzed for the presence of two classes of domains, and it is important for users to understand the difference between them. One class, called 3D Domains, is based solely on similar, compact substructures, whereas the second class, called Conserved Domains (CDs), is based solely on conserved sequence motifs. These two classifications often agree, because the compact substructures within a protein often correspond to domains joined by recombination in the evolutionary history of a protein. Note that CD links can be identified even when no 3D structures within a family are known. Moreover, 3D Domain links may also indicate relationships either to structures not included in CDD entries or to structures so distantly related that no significant similarity can be found by sequence comparisons.